Multilevel exploration of individual- and community-level factors contributing to overweight and obesity among reproductive-aged women: a pooled analysis of Bangladesh Demographic and Health Survey, 2004–2018

Objectives: Overweight and obesity have been related to a variety of adverse health outcomes. Understanding the overweight and obesity epidemic in Bangladesh, particularly among reproductive-aged women, is critical for monitoring and designing effective control measures. The purpose of this study was to determine the prevalence of overweight and obesity in reproductive-aged women and to identify the risk factors of overweight and obesity. Design: A total of 70 651 women were obtained from the five most recent and successive Bangladesh Demographic and Health Surveys (BDHS). The multilevel logistic regression model was used to explore the individual- and community-level factors of overweight and obesity. Setting: Five most recent nationally representative household surveys across all regions. Participants: Reproductive-aged (15–49 years) non-pregnant women. Results: Approximately 35·2 % (95 % CI: 34·9–35·6 %) of women were either overweight or obese in Bangladesh. At the individual- and community-level, higher age (adjusted odds ratio (aOR) = 5·79, 95 % CI: 5·28–6·34), secondary or higher education (aOR = 1·69 [1·60–1·78]), relatively wealthiest households (aOR = 4·41 [4·10–4·74]), electronic media access (aOR = 1·32 [1·26–1·37]) and community high literacy (aOR = 1·10 [1·04–1·15]) of women were significantly positively associated with being overweight or obese. Whereas, rural residents (aOR = 0·79 [0·76–0·82]) from larger-sized households (aOR = 0·80 [0·73–0·87]) and have high community employment (aOR = 0·92 [0·88–0·97]) were negatively associated with the probability of being overweight or obese. Conclusion: Individual- and community-level factors influenced the overweight and obesity of Bangladeshi reproductive-aged women. Interventions and a comprehensive public health plan aimed at identifying and addressing the growing burden of overweight and obesity should be a top focus.

haemorrhage and eclampsia (3,4) . Understanding the pattern of overweight and obesity and their determinants, especially among reproductive-aged women, is vital to secure the health of current and future generations. Consequently, the prevalence and determinants of overweight and obesity among reproductive-aged women are of particular interest to researchers and policy-makers. The rate of overweight and obesity is rapidly increasing in Bangladesh as a result of rapid urbanisation, economic growth, changes in dietary habits and lifestyle and modernisation (5) . The recent Bangladesh Demographic and Health Survey (BDHS) report showed that overweight and obesity among reproductive-aged women have increased from 12 % to 32 % between 2007 and 2017 (6) . Unlike undernutrition, overweight or obesity is a kind of malnutrition that receive little attention in Bangladesh (7) . Because of the rising prevalence of overweight and obesity in Bangladesh, it is necessary to take this issue into consideration while developing health and nutritional policy. Several studies have attempted to discover the associated factors of overweight and obesity in different settings using a single survey, including Bangladesh (8)(9)(10) . Individual factors associated with overweight and obesity have been discovered previously. However, the risk factors for overweight and obesity at the community level have received scant attention. A recent study looked at the prevalence and risk factors of overweight and obesity among Bangladeshi reproductiveaged women (11) . They determined the risk factors for underweight, overweight and obesity using pooled data from 1999 to 2014 and a multinomial logistic regression model. However, community-level factors were not considered in this study. Individual-level factors are deep setters in the community-level factors. Thus, disregarding community-level factors is a flaw of the previous research.
There is strong evidence that community factors, such as wealth status and literacy, have played a role in people being overweight and obese (7,12) . As a result, making inferences without considering individual-or community-level factors may result in erroneous information. This study aimed to investigate the individual-and community-level factors responsible for overweight and obesity among reproductive-aged women by conducting a multilevel analysis that considers the clustering and hierarchical structure of different survey years. Multilevel analysis has been utilised in a few research, although the results are outdated. The reason for this is that numerous changes in other socioeconomic elements have happened over time, including media campaigns using radio, television and educational status. Bangladesh has raised literacy rate (58·6 % to 72·3 %) and per capita income in recent years (13) . Women who live in highly educated societies and have access to the media are more aware of their own health and wellbeing (7) . The reason for this is that the public has access to nutritional information through the media (14) . As a result, the literacy rate of women in the community, wealth position, media access and employment may all have an impact on nutritional status and individual-level variables. Using the most recent nationally representative data from Bangladesh, this study examined the patterns and prevalence of overweight and obesity among reproductive-aged women and evaluated its association with individual-and community-level factors.

Data and sampling design
This study analysed data from five consecutive surveys undertaken by the BDHS in 2004, 2007, 2011, 2014 and 2017-2018. The Bangladesh Ministry of Health and Family Welfare and the National Institute of Population Research and Training conducted the survey. All BDHS employed a two-stage stratified cluster sampling procedure for selecting the sample. In the first stage, enumeration areas were randomly selected according to the probability proportional to the size of each enumeration area. In the second stage, the average number of households per enumeration area was chosen for each division to provide reliable sample estimates for the whole country. The surveys included 11 440, 10 996, 17 842, 17 863 and 20 250 married reproductive-aged women from five consecutive crosssectional household surveys. Missing values on the outcome and explanatory variables and women who were pregnant or had given birth within the previous 2 months of survey visits were excluded. A total of 70 651 women were evaluated for the subsample analyses after the exclusion criteria were met. The detailed information on the sampling techniques, survey design, survey instruments, measuring system, data collections and quality control has been previously described (6,(15)(16)(17)(18) .

Study variables
Outcome variables Overweight and obesity status was considered as the outcome variable and measured by the BMI of reproductive-aged non-pregnant women. The BMI cut-off points recommended by the WHO have been used to measure Bangladeshi women's overweight and obesity status (5,10,11) . According to the WHO, women are considered overweight when their BMI is 25·0 to 29·99 kg/m 2 , obese when their BMI is ≥ 30 kg/m 2 and overweight/ obese when their BMI is ≥ 25·0 kg/m 2 (19) . However, Bangladesh being an Asian country, this study used BMI cut-off point for Asians which is lower than the WHO criteria and categorised weight status into four groups as follows: underweight (< 18·5 kg/m 2 ), healthy weight (18·5-22·9 kg/m 2 ), overweight (23-27·5 kg/m 2 ) and obese (>27·5 kg/m 2 ) (20) . Finally, the primary outcome variable was formed by combining overweight and obesity for a woman with a BMI of ≥23 kg/m 2 .
Demographic and health survey data do not directly provide community-level characteristics except the place of residence (urban and rural) (19,21) . Therefore, other community-level factors were created by aggregating the individual-level factors inside their clusters. The distribution in each community is considered to develop community-level characteristics. This study used the mean or median as a cut-off point to create community-level factors. Current analyses also included four derived community-level variables, namely community poverty (low and high) (21) , community women literacy (low and high), community electronic media access (low and high) (21) and community women employment (low and high). Community-level factors such as community poverty, community women literacy, community electronic media access and community women employment were derived from the individuallevel factors, wealth index, women's education, electronic media access and women's working status, respectively. The detailed descriptions of the individual-and community-level factors are described in supplementary Table 1.

Statistical analysis
Descriptive analyses were employed to calculate the frequencies and percentages for each of the studied variables, followed by the estimation of the prevalence of overweight and obesity by the individual-and community-level factors. The Chi-square test was used to examine the differences between the group of overweight and obesity, and the significant (P < 0·05) variables for the Chi-square test were considered for multilevel analysis. Multilevel logistic regression modelling was applied to explore the relationships between individual-and community-level factors with the risk of being overweight or obese. Given that the individual-level factors were nested inside the community-level factors, a two-level multilevel modelling technique was used. Associations between individual-level predictors and overweight or obesity were measured by level 1 modelling, and the community-level determinants of overweight/obesity were assessed by level 2 modelling. Four separate regression models were run in this study. Without consideration for any individual-or communitylevel predictor variables, Model 1 is an empty model developed to estimate the random cluster effect and measures variation among the communities (primary sampling units).

Model 2 includes all selected individual-level factors in the bivariate analysis. Model 3 includes only community-level factors. The final model (Model 4) consists of individualand community-level factors.
To evaluate the best-fitted model, this study considered log-likelihood, Akaike's information criterion and Bayesian information criterion to explain the variation of these models. A 5 % significance level and a 95 % CI were used to measure the association between selected individual-and community-level factors and overweight/obesity. Multicollinearity was checked before establishing the final models. There is no multicollinearity identified because variance influential factor is less than the threshold value, 10. Therefore, multicollinearity diagnostic results were not reported in the paper. STATA 13.0 (StataCorp, USA) was used for all statistical analyses. Table 1 presents the distribution of overweight and obesity by individual-and community-level factors in Bangladesh. A total of 70 651 non-pregnant women were included. Among them, the highest percentage (35·1 %) was from the 20-29 years age group, 43·3 % of the women had secondary and a high level of education, and more than twothirds of the women were not working (68·9 %). The majority (89·6 %) of the women's religious faith were Islam. More than half of the participants had five and above household members (59·9 %) and 3-4 children (50·2 %), and 55·3 % of their families had electronic media access. At communitylevel factors, approximately two-thirds (64·1 %) of women were from rural areas and 52·9 % of women had high community-level electronic media access. Figure 1 presents the prevalence of overweight/obesity according to the WHO recommended BMI cut-off point and BMI cut-off point for the Asian population. The prevalence of overweight/obesity according to the WHO suggested BMI cut-off points was lower since it followed higher cut-off points than the BMI cut-off point for the Asian population. Fig. 2 also provides the prevalence of overweight and obesity according to the BMI cut-off point for the Asian population. The prevalence of overweight and obesity has increased in recent years. Table 1 provides the prevalence of overweight, obesity and overweight/obesity by individual-and communitylevel factors in Bangladesh. The overall prevalence of overweight and obesity were 24·3 % and 10·9 %, respectively. At the individual level, the prevalence of overweight/obesity has increased in recent years. The prevalence of overweight/obesity was the highest among the age group of 30-39 years (42·0 %) and lowest among the age group of 15-19 years (14·8 %). This study showed the highest rate of overweight/obesity among secondary and higher educated (44·3 %) women in Bangladesh. The prevalence of overweight/obesity increased with the wealth status of women. A small family (1-2) had a high prevalence of overweight/obesity (38·6 %). The prevalence of overweight/obesity was high among families with access to electronic media (43·7 %). Half (50·0 %) of women in the urban areas were overweight or obese at the community level. The prevalence of overweight/obesity was high among highly educated women in the community (38·8 %) and those with increased community electronic media exposure (39·4 %). All considered individual-and community-level factors except women's working status were significantly (P < 0·05) associated with overweight/obesity among women. Table 2  In the final full model (Model 4), individual-level factors (such as women's age, education, wealth index, household size, religion, electronic media access and marital status) and community-level factors (such as place of residence, community women literacy and community women employment) were significantly associated with overweight/obesity among non-pregnant women.

Discussions
This study draws required information from the five most recent BDHS data to assess individual-and communitylevel factors associated with overweight and obesity among non-pregnant reproductive-aged women in Bangladesh. Multivariate regression techniques were used to quantify the contribution of the individual-and community-level factors to overweight or obesity. Several studies used socioeconomic and demographic characteristics, and most of the reports applied a single model such as binary logistic regression. However, the investigation of communityand individual-level factors from the Bangladesh perspective is minimal. The findings of this study revealed that approximately one-third (35·2 %) of the non-pregnant women in Bangladesh were either overweight or obese. At the individual level, the prevalence of overweight/ obesity was high among women of the 30-39 years age group, wealthiest, married non-pregnant, with secondary and higher education, have 1-2 household members and with electronic media access. At the community level, the prevalence of overweight/obesity was high among non-pregnant women living in the urban area, from low community poverty, high community literacy and increased community electronic media access. In this study, the overall pooled prevalence of overweight/obesity in the nationally representative Bangladeshi non-pregnant women was 35·2 %, and the prevalence rates of overweight and obesity were 24·3 % and 10·9 %, respectively. These values are higher than those in Nepal (9) and Tanzania (22) . Therefore, overweight/obesity among women remains a serious concern in Bangladesh.
Several individual-and community-level factors affect the overweight/obesity of non-pregnant women. Individuallevel factors (such as women's age, education, wealth index, religion, household size, religion, number of children, electronic media access and marital status) and community-level factors (such as place of residence, community women literacy and community women employment) had a significant effect on overweight/obesity among non-pregnant women in Bangladesh. The risk of overweight/obesity increases with the women's age. Furthermore, the high age of non-pregnant women was significantly associated with higher odds of being overweight/obese compared with those in women of the low age group (15-19 years). An earlier study conducted in Bangladesh found that older women were at a high risk of being overweight/obese (23) . Another study conducted in India found that overweight and obesity increased significantly with age (24) . The fat mass may be the main reason for being overweight/obese. The results of another study suggested that older age is associated with considerable changes in body composition because of the gradual decrease in fat-free mass and increase in fat mass after 30 years of age (25) . Women's education has a positive relationship with the risk of being overweight/obese. Similar results have been reported in Bangladesh (23,26) and other countries such as Tanzania (22) , Nepal (9) , Malawi (19) and Ghana (27) . A possible reason could be that highly educated people lead a sedentary lifestyle (26) . Previous studies in low-and middle-income countries also revealed a significant positive relationship between education and overweight/obesity (28) . Even research conducted in developed countries found a negative relationship between education and excess body fat (29) .
Non-pregnant women from better-off households had higher odds of being overweight and obese than women from poorer households. This finding is consistent with a study conducted in low-to middle-income countries that found a significant positive association between wealth index and overweight/obesity (30) . A similar finding was also found in India (24) . The positive association between increased wealth and being overweight could be attributed to the dietary behaviour changes with high income. The intake of high energy density foods and consumption of animal and processed foods were high among the highincome households and significantly associated with overweight and obesity (31) . Upliftment in the wealth index has given great access to food and discharge from physical work to people in lower-income countries, leading to a high risk of being overweight and obese (11) . Additionally, the economically solvent and educated family often passes their time outside of the home due to job responsibilities and habituates to eating junk or fast foods outside. These activities may lead to an unhealthy life and increase the risk of overweight/obesity (32) . Furthermore, non-pregnant women whose families had access to electronic media were more likely to be overweight/obese than those whose families had no access to media. Television and radio are essential media sources, and women who watch television have a high risk of being overweight and obese (27,33,34) .
Mainly television watching has been used as a proxy for sitting time. Studies that followed participants over long periods have consistently found that people who spend more time watching television are likely to be overweight and obese (35) .
In the context of residence, women who lived in a rural area were less likely to be overweight/obese than those who lived in an urban area. The association between place of residence and overweight and obesity is evident in past studies conducted in India (36) , Senegal (37) and low-and middle-income countries (22,38) . A possible reason could be that rural women are highly engaged in physical activities such as agricultural activities, homework and other manual work as their occupation. These physical activities helped them reduce weight and hinder excessive weight gain (39) . Another possible reason could be that the availability of fast-food consumption, processed, packed and refrigerated foods is lower among rural women than among urban women (22) . Women from communities with high literacy were likely to be overweight/obese. However, high community-level literacy increases the risk of overweight and obesity (40) . Even health literacy is essential to understand consumer needs related to overweight and obesity (41) . An imminent investigation should discover the independent effects of community-level literacy on overweight and obesity among non-pregnant women in Bangladesh.
This study identified the individual-and communitylevel determinants of overweight and obesity among non-pregnant women aged 15-49 years in Bangladesh. The findings have important policy implications. The use of multilevel analysis helped solve practical and methodological constraints. Owing to the use of the nationally representative dataset, the findings can also be generalised among non-pregnant women.
The key strength of this study is its extensive pooled data from five successive BDHS rather than focusing on a specific survey year. Thus, the study is more diverse, and the results are more cohesive than any cross-sectional studies. This study identified important individual-and community-level factors predicting overweight and obesity among reproductive-aged women in Bangladesh. It has described community-level variables within a multilevel model framework. Therefore, unlike studies with only individual factors, it has minimised omitted variable biases. Furthermore, the results are based on five successive nationally representative surveys with large sample sizes, making the results generalisable to other similar socio-economic settings.
This study has also some limitations. The analysis was based on the secondary data, and some important variables, such as food habits, smoking, smoking habits, some non-communicable diseases risk and physical activity of women, were not included due to unavailability. The inclusion of these variables in the models might be helpful to fully understand the relationship of the selected independent variables with the overweight and obesity status of women. This study followed a cross-sectional design that cannot provide a causal relationship between explanatory variables and overweight and obesity. Further, the analysis was limited to only non-pregnant women aged 15-49 years; thus, the generalisability of the findings to all women may be limited. Despite these limitations, the results significantly contribute to the existing literature on the association of individual-and community-level factors with women's overweight and obesity status.

Conclusions
Individual-and community-level factors of overweight and obesity among reproductive-aged women in Bangladesh were investigated in this study. The study found that the prevalence and likelihood of being overweight/obese sharply increased over the survey period. At the individual level, women's age, women's education, wealth index, religion, household size, numbers of children and electronic media exposure have a significant relationship with overweight/obesity. At the community level, place of residence and community women literacy are significantly associated with a higher likelihood of being overweight/obese. Older, educated and women from the wealthiest households should be conscious about overweight/obesity and its harmful effects on health and quality of life. Media exposure should play an essential role in providing health awareness messages across all communities. Finally, executing protective interventions by government and nongovernment organisations may reduce the growing burden of overweight/obesity, particularly among the reproductiveaged women in Bangladesh. and Health Surveys (DHS). We acknowledge the DHS for providing us this freely available data. Financial support: This research did not receive any specific grant from funding agencies in the public, commercial or not-for-profit sectors.